How might software development have unfolded if CPU speeds were 20x slower?

Hacker News Discussion ↗

Overall framing

  • Many argue this isn’t very hypothetical: CPUs really were ~20x slower 10–20 years ago, so looking back is instructive.
  • Others push back that the question assumes hardware stays slow while software keeps evolving, which is different from “we’re just 8.6 years earlier on Moore’s law.”

Performance, optimization, and bloat

  • Consensus that tighter CPU budgets would force:
    • More attention to algorithmic efficiency, memory usage, and cache behavior.
    • Less tolerance for heavy abstraction layers and “just throw hardware at it” attitudes.
  • Several note that “doing less” is usually the biggest optimization; 20x slower hardware would reinforce this.
  • Wirth’s Law and Jevons-like effects are invoked: faster hardware has been largely “eaten” by heavier software stacks.

UI frameworks, desktop vs web

  • Many predict:
    • More native C/C++ desktop apps, fewer large browser engines and Electron-style apps.
    • UI frameworks more like VB6/Delphi/early Qt/WPF, emphasizing fast, small binaries and responsive controls.
  • Strong nostalgia for 90s/2000s RAD tools and native toolkits; people contrast them with sluggish, multi-layer web UIs.
  • Others note old systems were often sluggish on their contemporary hardware; perceived “speed” today is partly running old software on new machines.

Parallelism, GPUs, and architecture

  • Several expect much heavier emphasis on:
    • Multicore, SIMD, and GPGPU earlier and more pervasively.
    • Domain-specific accelerators and Cray-like vector/array processing.
  • Some discussion that slower clocks would also affect buses, memory, and displays, making the tradeoffs more complex.

Applications: games, web, AI, cryptography

  • Games: predictions range from “still Doom/Quake-level graphics” to “more strategic/intellectual designs and less flashy spectacle.”
  • Web: likely more static pages, fewer heavyweight client-side frameworks; less tracking/analytics overhead simply because it would be too slow.
  • AI: many doubt current large-scale deep learning and LLMs would exist; compute and training costs would be prohibitive.
  • Crypto: 20x slower/faster alone is minor for security margins; algorithms like 3DES/AES choices would change slowly.

Economics, culture, and skills

  • Slower hardware is seen as:
    • Increasing the value of deep systems knowledge, complexity analysis, and formal methods.
    • Reducing “bootcamp-style” shallow training and easy patch-after-release practices.
  • Some argue management would still prioritize shipping features over optimization; constraints help, but culture remains decisive.